Graph views via property maps

Hi all,

gt allows creating graph views where vertices and/or edges are filtered out. The principle is described in the documentation:

"Vertices or edges which are to be filtered should be marked with a PropertyMap with value type bool, and then set with set_vertex_filter() or set_edge_filter() methods. By default, vertex or edges with value '1' are kept in the graphs, and those with value '0' are filtered out." (

Given a layered graph, where layers are coded via an edge property, I want to draw the graphs layer by layer.

This is an example graph:
from graph_tool.all import *
import numpy as np
import pandas as pd
# data
edge_list = pd.DataFrame(np.array([[0, 1, 0], [1, 2, 1], [2, 0, 2]]), columns=['i', 'j', 'layer'])
# create graph
g = Graph(directed=False)
ep_layer = g.new_edge_property('int')
g.add_edge_list(edge_list.values, eprops=[ep_layer])
g.edge_properties['layer'] = ep_layer
# draw graph with edge colors showing the layers
graph_draw(g, edge_color=g.ep.layer)

It is possible to draw a single layer (in this example layer 0) in this complicated way:

# set layer to be drawn
layer = 0
# create graph view
ep_filter = g.new_edge_property('bool')
for i in range(0, len(edge_list)):
    e = g.edge(edge_list['i'][i], edge_list['j'][i])
    if edge_list['layer'][i] == layer:
        ep_filter[e] = True
        ep_filter[e] = False
g_filter = GraphView(g, efilt=ep_filter)
# draw filtered graph
graph_draw(g_filter, edge_color=g.ep.layer)

If I proceed this way I will also have to create a vertex property map to filter unused vertices and do these steps for all layers I want to draw.

But isn't there a more elegant way - preferably handling vertices and edges in the same step?

Many thanks and best wishes


attachment.html (6.37 KB)

Ni! Hi Haiko,

GraphView can simultaneously take `vfilt` and `efilt` parameters, if that
is what you wanna know. So,

g_filter = GraphView(g, vfilt=vp_filter, efilt=ep_filter)

should work just fine.

Regarding your code snippet, the 'elegance' issues are kinda unrelated to
graph_tool. For example, in the second part, you iterate the dataframe
instead of simply iterating the graph edges and getting the edge layer from
the internal property map. You also use an if clause to assign the boolean
value of the if clause, instead of directly assigning it. So, this is
equivalent to your `for` loop:

for e in g.edges():

    ep_filter[e] = g.edge_properties['layer'][e] == layer

All the best,


attachment.html (6.88 KB)

Hi Alexandre,

Thx for your reply! Indeed, your and my following optimizations are not gt-specific. I’m now here:

# set layer to be drawn

layer = 0

# create graph view

v_filter = edge_list.groupby('layer')[['i', 'j']].apply(lambda x: set(x.values.flatten())).to_dict() # create dictionary with vertices in a layer

vp_filter = g.new_vp('bool')

vp_filter.a[list(v_filter[layer])] = True

ep_filter = g.new_ep('bool')

ep_filter.a[g.ep['layer'].a == layer] = True # replaces my or Alexandre's loop

g_filter = GraphView(g, vfilt=vp_filter, efilt=ep_filter)

# draw filtered graph

graph_draw(g_filter, edge_color=g.ep.layer)

This is already rather short and nice. Still, I'm wondering if there’s a way that is more gt-esque.


attachment.html (16.1 KB)

You seem to be insisting on keeping an external list of edges for each
layer, and most of the complication stems from having to project from
this external data structure.

By far the simplest approach is to have a single edge property map with
the layer labels, and use that to do the filtering, i.e.

    elayer = g.new_ep("int")

    # populate g and elayer

    g_l = GraphView(g, efilt=elayer.fa == l) # layer l